Joint bunker fuel and freight revenue management in liner shipping: A decision-focused learning approach

The volatility of bunker fuel prices significantly impacts the maritime industry due to increasing regulatory complexities and market uncertainties. This study focuses on addressing these challenges by identifying bunker price prediction as a critical area of impact for improving operational efficiency and decision robustness. A comprehensive optimization model has been proposed to determine optimal ship speed and bunkering strategies both within and outside Emission Control Areas (ECAs). The traditional two-stage method involves training predictive models for bunker price, with the predictions being used as input parameters to solve the optimization problem. However, the loss function in this two-stage method does not consider the effect of predictions on the downstream decision-making problem. Therefore, this study adopts an integrated framework, Decision-Focused Learning (DFL) that unifies prediction and optimization processes into a cohesive approach. Unlike traditional sequential approaches, this method embeds the predictive model directly within the optimization process. It is comparatively more capable of handling data-driven optimization problems than traditional two-stage methods. In this study, computational experiments show that how DFL outperforms traditional methods, achieving 1.29% lower Normalized Regret, which translates to predictions resulting in approximately $22,730 higher profit for liner services. These findings offer valuable managerial insights into improving efficiency and sustainability in liner shipping operations by making better decisions under real world uncertainties.